Open source & no lock-in

Nothing built on polite.ai is trapped here. The platform core is open, configurations are plain data available for download, and the same API the dashboard uses is documented for direct calls.

The open-source core

The core that answers calls — the agent runtime and its API — is open source under the MIT licence at github.com/aplisay/llm-agent. Anyone can read exactly how the agents run, or run the core themselves. polite.ai is the hosted service built on that core: the Builder, the knowledge system, numbers, billing, and the dashboard around them — the fee pays for running it well, not for the absence of an alternative.

The Builder is a convenience, not a cage

Everything the Builder writes is an ordinary agent configuration. Any agent opened in the agent editor shows it all, editable by hand — prompts, models and voices, functions, API keys and MCP servers. There is no hidden layer. Underneath both sits the platform API — the same one the dashboard calls — documented in the Aplisay developer docs.

Export everything

• Download config — any team or agent's full configuration as JSON, from the team card's menu, the Builder header, or the agent drawer's Overview tab. That file is the agent: prompts, wiring, settings.
• Calls → Export CSV — call history for any date range.
• Usage → Export — raw metered usage as CSV (Usage analytics).

Numbers aren't hostages

A purchased number can be deleted at any time — it is released back at the carrier immediately. Numbers an organisation brings itself never left home in the first place: the carrier keeps them and routes calls over a private trunk, or an existing PBX connects outward via SIP registrations. Walking away is a routing change, not a migration project.

Build from another AI assistant

Aplisay publishes an MCP server that lets any AI assistant design, create and manage agents through the platform API — setup and details in the Aplisay developer docs.